Decoder takes on AI's messy climate footprint

AI has a massive power appetite, and it is easy to make it the ultimate villain in today's climate crisis.

But if you actually sit down to look at the tech world's energy ledger, the problem is a whole lot messier than just pointing fingers at chatbots.

On a recent episode of The Verge's Decoder podcast, host Nilay Patel sat down with senior science reporter Justine Calma to untangle this exact tension.

The conversation cuts straight to the dilemma: our aging energy grids are struggling to keep up, yet big tech is racing to burn through power faster than ever.

Big tech's climate targets go out the window

Running frontier AI at scale means building massive data centers packed to the brim with power-hungry GPUs.

That reality has derailed years of corporate promises. Tech giants like Microsoft, Google, and Amazon spent years working with governments on ambitious sustainability goals, trying to keep climate change from turning into a full-blown catastrophe.

Now, the AI boom has blown those roadmaps wide open. These companies are not just missing their emissions targets; their numbers are actively getting worse over time.

It sparks two big questions. The practical one: can our fragile power grid actually handle this sudden spike? And the moral one: should we even be building these systems if they accelerate environmental damage?

It is not just an AI problem

Here is the twist: packing servers into data centers and running them at full throttle is already how modern tech operates.

If you object to AI purely on energy grounds, the same moral dilemma applies to a lot of things we use every single day.

Take TikTok and YouTube, which constantly ingest and transcode millions of hours of video around the clock. Or video games, which push power-hungry GPUs in living rooms and demand heavy server power for online multiplayer. And crypto mining faces even harsher criticism for similar reasons.

Patel highlighted a sharp hardware comparison: the Nvidia H100, the crown jewel of AI training, pulls roughly the same amount of power as the gaming-focused Nvidia RTX 4090.

If the electricity draw is basically identical, what framework should society use to decide which use of that power is actually justified?

Why it matters

We think this conversation matters because it strips away performative outrage and focuses on the real tradeoff: deciding what kind of computation is genuinely worth the juice.

For regular users, telling people to stop using AI makes no more sense than telling the world to stop streaming video or playing games. Every digital convenience carries an invisible carbon cost.

For the tech industry, the pressure is mounting. The old playbook of offsetting emissions on paper will not hold up when physical power grids hit their limits. The energy crunch is quickly becoming a hard bottleneck on pure computing power.

AI is not the sole cause of the climate crisis, but it is forcing an overdue reckoning. The question is no longer whether we can afford the power, but whether tech companies will finally be forced to invent their way toward real energy efficiency.